Automated trading systems cannot work. Discuss!

[Medic]

Seriously. If it were possible to script indicators into strategies to reliably enter winning trades then stock market actions would be dominated by robots in the same way that online poker is.

So many people selling their 'winning' system doesn't make any sense. If the system was successful the author would set it going and live happily ever after.

With so many market factors/parameters and mass croud actions of millions of traders it must be impossible to develop a rule based strategy for winning. On the other hand there is roughly the same number of car crashes every year which reveals the underlying order of human behaviour.

Love to hear comments from both sides on this?

[ctrlbrk]

Hello Medic,

Welcome to the NinjaTrader Forums!

About a week or two ago I blogged about this and there was a pretty lively discussion on my blog from users. I used to be "heavy" into automated strategies as the only way to fly, but to summarize the blog post, I am now focusing on discretionary setups and use my programming skills to assist my discretionary trading instead of put it on auto-pilot.

Mike

[Operandi]

I know from experience that some automated trading systems do work. For example, there are NT customers on this forum who say that they have developed winning systems.

Of course I don't know if their systems win overall in all market conditions, which is obviously critical for a long term automation model.

In general, though, traders rely too heavily on indicators which muddy trading decisions with delays and distortions. Whilst this is fine in discretionary trading where the trader can decide to take or leave a signal through experience and knowledge, it can turn a successful manual system into a failing automated robot.

The most successful systems that I come across get as close to the price action, and/or the dynamics of volume or the market, as possible.

[Operandi]

I've thought about this a bit more. To me the following chart sums up the frustrations and challenges, and rewards (!) of an automated trading system.

At support each acceleration looks very similar, however resistance was hit causing ranging action. Most trend following automated systems would lose money at this point as they would be too slow to react to the rapid changes in direction.

Detecting that breakout and rejected the prior ranging action leads to significant gains. If your system can do that and then get out in time to lock in your profits, then that one trade is all you need each day.

Once it succeeds in that one that trade then program it to stop for the day.

Sounds simple in practise, but that is the challenge. What caused that ranging action? What caused that breakout? How can you detect and manage those market dynamics?

[ctrlbrk]

For me, the biggest strength of automation was to eliminate the emotional and mental aspect of trading. Many times when I lose money on a trade, it was a direct result of me not following my own predefined rules.

Automated strategies are excellent at following the rules.

However. I stopped devoting all my time and energy to automation because I realized that there was trade after trade I was missing that I would have taken discretionary, but for one reason or another the strategy did not. Usually, the reason was because the backtest said it was not profitable in the long run, yet every chart I could stand to stare at said it worked.

So. For me, all the time and energy of automation was to try to cure my mental woes. Today I am taking a different approach to fix that problem, although I still desire to have a functioning well-oiled strategy that I can depend on daily to make money.

Speaking of strategies, you may wish to check out bluewavetrading.com. Randy just released his new Ninja auto trader. It is expensive -- $ 3500 per quarter for a limit of 5 contracts and you must own the BWT indicators, which is another $ 3500 (lifetime). So $ 7000 if you are new to BWT. But, he claims it is making money hand over fist (made $3000 last week I believe).

Mike

[ctrlbrk]

Operandi,

I agree with your approach of stopping after hitting your daily goal. No question that the more trades you make, the more risk there is. The market should be treated with extreme caution to "get in, get out" and do it a minimal number of times to improve your overall odds of being successful.

I even developed several strategies that only traded once a day, maybe twice, looking for that perfect setup all the while trying not to curve fit the optimizations. It is very difficult to feel good about the results unless you can backtest for at least a year. I want at minimum 100-200 trades in my backtest report and 1 year of period before I trust it is not curve fitted.

This is where NT 6.5 is the deal-killer, the inability to backtest any large periods of tick, range, volume data. Not to mention the whole contract rollover PITA (for Zen users especially). So, I hope NT7 will fix both of those and make long term tick-level backtesting a reality, which should help cure the "curve fitting blues".

Mike

[ctrlbrk]

Sorry for the extra posts. Operandi - for that range bound area, check out Murrey Math lines, they plotted that range on Friday in advance.

Mike

[DangerBoy]

Hi guys

Its funny but it's the second time today that I've heard someone say that you should stop trading once you hit a certain amount for the day. A chap in my trading room mentioned it earlier with a scalping system where you only take those few trades per day and once you've made your money you stop.

Personally I can't see any logic in this idea, I understand the less time in the market the less exposed you are to risk but it seems to me that you are just lengthening out the amount of time it will take to actually stumble across all the variations. For instance if i do 20 scalps a day over a week, I'm am much more likely to hit more variations and in a short period of time, I will find out fairly quickly if my system is a winner or not.

If I take the same system and do one trade a day I am still going to hit all the variations but it just takes longer for me to hit them and so deceptively a system may look to be winning in the short run but you could get killed the minute you eventually hit some price action that is not favourable for your system.

I would love to hear the rationale for stopping trading once you make a certain amount for the day.

Cheers
Ross

[Medic]

Hmmm interesting.

I write pattern recognition for medical EKG systems and have started applying my skills to the market. First two trades resulted in 56% and 38% return (over 3wks). Third trade took a 12% loss.

I'm learning quick as I'm trading real money from the start. It occurs to me that all I've read is just the tip of the iceberg and even that is what people have figured out so far.

I'll keep going until my trades are all upside or I go broke :-)

Any good tuts on price action greatly apprecaited.

Thanks.

[Operandi]

Hi Medic

As someone with a background in physiology, including ECG, EEG and sleep analysis systems myself, your ability to isolate signals from noise will probably be valuable. It'll be interesting to hear how you get on with market analysis.

With regards to the over trading point, systems can over trade just as much as humans can (and probably more so). Good profits can dwindle away in periods of chop and sudden reversals can destroy your profits for the day.

To me, a system that waits for the high probablity setup, trades it effectively and then shuts itself down once it had harvested good profits, is the holy grail.

To detect a high probablity trade, but also reject weaker setups is not easy. Having that system hit one or two 5 point runs on the ES each day and then shut down, would probably make as much money as you need as you can ramp up the number of contracts to achieve scale.

[Medic]

Hi Operandi,

At the moment I'm trading stocks whilst I look for a broker for E-minis. I played with Zen-Fire through Mirus and was less successful using pure technical analysis to understand the market. With my knowleadge of medical and FDA processes I first select stock by company development plans and fundementals then use a stragety on that stock. Swithing to S&P or DOW E-Mini would lose much of this edge...

still more learning than losing for this week. The show goes on.

[-Swig-]

Operandi,

This is where NT 6.5 is the deal-killer, the inability to backtest any large periods of tick, range, volume data. Not to mention the whole contract rollover PITA (for Zen users especially). So, I hope NT7 will fix both of those and make long term tick-level backtesting a reality, which should help cure the "curve fitting blues".

Mike

ctrlbrk, what's stopping you from backtesting large periods? Memory problems?

[ctrlbrk]

Contract rollover issues (don't get me started) and memory issues. With the largeaddress patch the memory issues are much more under control now so that is great.

Still, we need a Genetic Optimizer built-in so we can do some smart backtesting. And how about displaying the best performance value found so far real-time, so I can see if an acceptable number has been found in the first 5 seconds instead of waiting 50 minutes or 5 hours etc.

But I digress. I am focused on discretionary right now, I have set aside all aspirations to find an automated trading strategy that is profitable for the long haul. And I don't say this lightly or without a lot of experience to back up my findings.

Mike

[DangerBoy]

ctrlbrk,

I see your posts quite often when I'm looking around here and read them with interest. I see you are moving back to discretionary and using your coding skills as an extra tool.

What are your thoughts on having an automated system that trades as a sort of framework from within which you trade discretionary? My system often trades well but sometimes there are points when I think, ah that trade just shouldn't be taken or should be closed but I can't meddle with it for fear of the strategy getting out of align with where it should be.

Have you tried this at all, trying to change the strat to look at an external reference point that perhaps you could modify yourself, like an external points based system with default values, you can manually adjust these points higher or lower and the strat will take a different course of action on the next bar update if it sees a certain parameter has a point score that has gone beyond a predefined threshold.

Just a thought, interested to hear your comments since you have abviously been there and tried most things.

Ross

[ctrlbrk]

Ross,

I've tried everything I could think of. And things other people could think of. It would take me a novel to just describe an overview of everything I have tried. Much less the ba-zillion lines of code I have written trying to make them work.

Don't get me wrong. Many of them made a killing. For a while. Then bam.

I am of firm belief your strategy needs to be backtested for at minimum 1 year, probably more, and contain at minimum 100 trades, preferably many many more (200-500). If it is profitable after all that, congratulations, you've made it to step 1 of 10. Now the hard part is having it continue to make money over the next year.

You can tell Ninja to sync the strategy in a smarter way by going to Options -> Strategies -> NnjaScript, and selecting "Wait until flat". Why that isn't default I do not know. Once I found that option I sighed with relief, hopefully it helps you too.

Roonius from TradingStudies also has a great tool for adding discretionary buttons to your chart toolbar. These buttons could turn a strategy on or off (ie during chop). Or maybe you only want longs, or only want shorts. Great stuff.

The closest I came to doing what you are calling an external point system, is to create a Score based system where many, many factors are considered in determing the "score" of a trade. Multiple time frames, support, resistance, trend, yada yada blah blah.

I have not done something where I manually weigh the market and place these weighted values somewhere and have the strategy look at them to decide to trade (in real time). I think it might be better/easier/simpler to use Roonius' buttons to do that type of thing (ie "only longs", "only shorts"). Perhaps a button for "wait for long failure, then go short" type of thing.

I have coded strategies that pull data from a MySQL database to try to adapt to the market, but that's a whole different thread.

Mike